AI Mammography Risk Scores Rise Years Before Breast Cancer Diagnosis

In a study of 54,014 women, an AI deep learning model's 5-year breast cancer risk score increased from a median of 2.1 to 6.6 in the 6 years before a cancer diagnosis. Scores for women who remained cancer-free were stable (range, 1.8-2.2), suggesting AI risk scores could act as…

• Multisite retrospective cohort of 158,807 screening mammograms from 54,014 women (817 with cancer); used linear mixed-effects models to assess score trajectories.
• In the cancer group, longitudinal risk scores significantly increased by 1.13 per year [95% CI: 1.07, 1.18], while the cancer-free group had a minimal annual change of 0.09 per year.
• This was a retrospective analysis, so the findings require prospective validation before clinical implementation for risk-adaptive screening.

RadPigeon summaries are original and for information only. They are not clinical advice.